CS 6501 Computational Visual Recognition Final Day Outline

CS 6501: Computational Visual Recognition Final Day

Outline for Today • Recurrent Neural Networks Lab Recap • Course Recap • Ethics and Scholarship in AI / Vision CS 6501: Computational Visual Recognition 2

Recurrent Neural Network Cell CS 6501: Computational Visual Recognition

Recurrent Neural Network Cell CS 6501: Computational Visual Recognition

Recurrent Neural Network Cell CS 6501: Computational Visual Recognition

Recurrent Neural Network Cell CS 6501: Computational Visual Recognition

Recurrent Neural Network Cell e (0. 7) abcde CS 6501: Computational Visual Recognition

Generating Samples from the Recurrent Neural Network Cell CS 6501: Computational Visual Recognition

LSTM Cell (Long Short-Term Memory) CS 6501: Computational Visual Recognition

How do we train the network? We don’t do it character by character. CS 6501: Computational Visual Recognition

Recurrent Neural Network Cell CS 6501: Computational Visual Recognition

Recurrent Neural Network Cell CS 6501: Computational Visual Recognition

(Unrolled) Recurrent Neural Network a t <<space>> c a t CS 6501: Computational Visual Recognition

(Unrolled) Recurrent Neural Network cat likes the cat CS 6501: Computational Visual Recognition eating likes

(Unrolled) Recurrent Neural Network positive / negative sentiment rating the CS 6501: Computational Visual Recognition cat likes

(Unrolled) Recurrent Neural Network a t c a CS 6501: Computational Visual Recognition <<space>> t

CS 6501: Computational Visual Recognition

Bidirectional Recurrent Neural Network gato quiere the cat CS 6501: Computational Visual Recognition comer wants

Stacked Recurrent Neural Network c a CS 6501: Computational Visual Recognition t

Bidirectional Stacked Recurrent Neural Network c a CS 6501: Computational Visual Recognition t

CS 6501: Computational Visual Recognition

Course Recap (1) • Convolutions / Filtering Images / Blur / Edges / etc • Color Spaces HSV / Saturation Enhancement • Linear Classifier (Softmax) • Softmax Loss / Gradient Computation • Training / Validation / Test • Stochastic Gradient Descent CS 6501: Computational Visual Recognition

Course Recap (2) • Convolutional Neural Networks / Backpropagation • Alex. Net, VGG, Goog. Lenet • CIFAR-10 Dataset (Obtained 80% Accuracy, and some of you 90%) • Use a pretrained Convnet to obtain Features. • Repurpose a pretrained Convnet for another task. CS 6501: Computational Visual Recognition

Course Recap (3) • Object Detection (RCNN, Fast. RCNN, YOLO) • Image Segmentation (Fully Convolutional Networks) • Object Proposals (Box Proposals, Segment Proposals) • Generative Adversarial Networks (GAN) / Deep Dream • Siamese Networks (For Comparing Images / Patches) • Visual Recognition for Videos • Place / Scene / Location Recognition CS 6501: Computational Visual Recognition

Course Recap (4) • Inception Network / Residual Network • How to make networks faster? XNORNet, EIE. • Principles of Categorization / Intuitions from Psychology • Recurrent Neural Networks (Image Captioning, Text Generation) • And finally, you hopefully are proficient now with either Torch (Facebook fans) or Keras/Tensorflow (Google fans). CS 6501: Computational Visual Recognition

Scholarship and Ethics (What not to do) [In class discussion] CS 6501: Computational Visual Recognition

Scholarship and Ethics (What not to do) [In class discussion] CS 6501: Computational Visual Recognition

Scholarship and Ethics (What not to do) [In class discussion] CS 6501: Computational Visual Recognition

Reminder about your Projects Presentations are 10 minutes total. * Project report is due on December 5 th for everyone. * Use inspirations from the papers we have read during class to elaborate your final report. * Make sure you include an introduction motivating your problem (probably similar to what you already wrote in your proposal and progress report) * Make sure you also include a clear description of your method, including figures of the model if necessary, a clear description of your algorithm and parameter choices, and discussion of your choices in the method. * Make sure you include figures, and plots, and tables and numbers of your experiments. * IMPORTANT: Make sure you include actual outputs of your method. For instance actual input images and outputs/predictions of your algorithm, including cases where it worked well, and cases where it might have failed. Include discussions of why it might have failed and what you could have done to improve those cases, etc. Trust your judgment for this part but again. take inspiration on the papers we have read this semester. Best luck to all! CS 6501: Computational Visual Recognition

Hope you enjoyed the Class. Thanks! CS 6501: Computational Visual Recognition